genai-semantic-normalizer

An OpenTelemetry Collector processor that normalizes vendor-specific LLM attributes to the standard gen_ai.* semantic conventions.

What It Does

LLM providers use different attribute names for the same concepts. This processor maps them all to the OpenTelemetry GenAI semantic conventions:

Vendor Attribute	Normalized To
openai.model	gen_ai.request.model
anthropic.input_tokens	gen_ai.usage.input_tokens
cohere.model_id	gen_ai.request.model
llm.model	gen_ai.request.model

Supports: OpenAI, Anthropic, Cohere, Azure OpenAI, Google/Vertex AI, and generic llm.* attributes.

Configuration

processors:
  genai_semantic_normalizer:
    enable_defaults: true     # Use built-in vendor mappings
    overwrite: false          # Don't overwrite existing gen_ai.* attrs
    drop_original: false      # Keep vendor-specific attrs after normalization
    custom_mappings:          # Add your own mappings
      my_vendor.model: gen_ai.request.model

Part of the AIR Platform

This processor is one component of the AIR Blackbox Gateway collector pipeline.

License

Apache-2.0